Hardware
Hardware in AI security refers to the physical parts—like the computer chips that power smart software—that can be tampered with by bad actors during manufacturing or shipping to compromise the system's safety.
Hardware security in AI involves protecting the physical compute infrastructure, such as GPUs, TPUs, and CPUs, from supply chain attacks where malicious modifications or backdoors are introduced to manipulate model execution or data integrity.
Hardware-level AI security encompasses the mitigation of supply chain vulnerabilities within the compute stack, including GPUs, TPUs, and embedded accelerators, where adversaries may introduce hardware trojans or side-channel vectors to compromise the confidentiality, integrity, or availability of AI model inference and training processes.